V2Recommendation
One (date, room_type) recommendation produced by the RL policy.
Possible values: Value must match regular expression ^(?!^[-+.]*$)[+-]?0*\d*\.?\d*$
Percent diff vs the room's current rate at compute time.
Possible values: Value must match regular expression ^(?!^[-+.]*$)[+-]?0*\d*\.?\d*$
How much of this date's input data was present, in [0, 1]. Not a measure of how sure the model is about the rate — a fully-fed date the engine has little to say about still reports 1.0.
Whether this row is a recommendation at all. PRICED is the only value carrying a rate change to act on. HELD_LOW_CONFIDENCE and INSUFFICIENT_DATA both return your current rate with delta_pct 0 — the row exists to say what happened to the date, not to be published. Check this before acting on recommended_rate: a hold is not an instruction to keep the rate, it is the absence of an instruction.
Possible values: [PRICED, INSUFFICIENT_DATA, HELD_LOW_CONFIDENCE]
PRICEDDemand reading for the date: DISTRESSED | BUILD | YIELD | SPIKE, or NOT_MEASURED where too little is on the books for ahead or behind to mean anything — most of the far horizon. Treat it as an open string: NOT_MEASURED is the absence of a reading rather than a fifth kind of demand, and further values may be added.
drivers object[]
Feature name (e.g. pace_index).
Humanised feature name for display.
Contribution magnitude.
up | down.
guardrails object[]
Machine-readable guardrail code.
Humanised label for display.
falseexplanation object
- string
- null
{
"date": "2024-07-29",
"room_type": "string",
"recommended_rate": "string",
"delta_pct": "string",
"data_completeness": 0,
"status": "PRICED",
"phase": "string",
"drivers": [
{
"feature": "string",
"label": "string",
"impact": 0,
"direction": "string"
}
],
"guardrails": [
{
"code": "string",
"label": "string"
}
],
"hitl_required": false,
"explanation": "string"
}